help graph bar dialog: graph bar
-------------------------------------------------------------------------------
Title
[G] graph bar -- Bar charts
Syntax
graph bar yvars [if] [in] [weight] [, options]
graph hbar yvars [if] [in] [weight] [, options]
where yvars is
(asis) varlist
or is
[(stat)] varname [[(stat)] ...]
[(stat)] varlist [[(stat)] ...]
[(stat)] [name=]varname [...] [[(stat)] ...]
where stat may be any of
mean median p1 p2 ... p99 sum count min max
or
any of the other stats defined in [D] collapse
mean is the default. p1 means 1st percentile, p2 second, and so on;
p50 means the same as median. count means the number of nonmissing
values of the specified variable.
options description
---------------------------------------------------------------------
group_options groups over which bars are drawn
yvar_options variables that are the bars
lookofbar_options how the bars look
legending_options how yvars are labeled
axis_options how the numerical y axis is labeled
title_and_other_options titles, added text, aspect ratio, etc.
---------------------------------------------------------------------
group_options description
---------------------------------------------------------------------
over(varname[, over_subopts]) categories; option may be repeated
nofill omit empty categories
missing keep missing value as category
allcategories include all categories in the dataset
---------------------------------------------------------------------
yvar_options description
---------------------------------------------------------------------
ascategory treat yvars as first over() group
asyvars treat first over() group as yvars
percentages show percentages within yvars
stack stack the yvar bars
cw calculate yvar statistics omitting
missing values of any yvar
---------------------------------------------------------------------
lookofbar_options description
---------------------------------------------------------------------
outergap([*]#) gap between edge and first bar and
between last bar and edge
bargap(#) gap between yvar bars; default is 0
intensity([*]#) intensity of fill
lintensity([*]#) intensity of outline
pcycle(#) bar styles before pstyles recycle
bar(#, barlook_options) look of #th yvar bar
---------------------------------------------------------------------
legending_options description
---------------------------------------------------------------------
legend_option control of yvar legend
nolabel use yvar names, not labels, in legend
yvaroptions(over_subopts) over_subopts for yvars; seldom
specified
showyvars label yvars on x axis; seldom
specified
blabel(...) add labels to bars
---------------------------------------------------------------------
axis_options description
---------------------------------------------------------------------
yalternate put numerical y axis on right (top)
xalternate put categorical x axis on top (right)
exclude0 do not force y axis to include 0
yreverse reverse y axis
axis_scale_options y-axis scaling and look
axis_label_options y-axis labeling
ytitle(...) y-axis titling
---------------------------------------------------------------------
title_and_other_options description
---------------------------------------------------------------------
text(...) add text on graph; x range [0,100]
yline(...) add y lines to graph
aspect_option constrain aspect ratio of plot region
std_options titles, graph size, saving to disk
by(varlist, ...) repeat for subgroups
---------------------------------------------------------------------
The over_subopts -- used in over(varname, over_subopts) and, on rare
occasion, in yvaroptions(over_subopts) -- are
over_subopts description
---------------------------------------------------------------------
relabel(# "text" ...) change axis labels
label(cat_axis_label_options) rendition of labels
axis(cat_axis_line_options) rendition of axis line
gap([*]#) gap between bars within over()
category
sort(varname) put bars in prespecified order
sort(#) put bars in height order
sort((stat) varname) put bars in derived order
descending reverse default or specified bar order
reverse reverse scale to run from max to min
---------------------------------------------------------------------
aweights, fweights, and pweights are allowed; see weight and see note
concerning weights in [D] collapse.
Menu
Graphics > Bar chart
Description
graph bar draws vertical bar charts. In a vertical bar chart, the y axis
is numerical, and the x axis is categorical.
. graph bar (mean) numeric_var, over(cat_var)
y
| numeric_var must be numeric;
7 | +--+ statistics of it are shown on
| | | the y axis.
5 | +--+ | |
| | | | | cat_var may be numeric or string;
| | | | | it is shown on the categorical
| | | | | x axis.
+------------------------- x
first second ...
group group
graph hbar draws horizontal bar charts. In a horizontal bar chart, the
numerical axis is still called the y axis, and the categorical axis is
still called the x axis, but y is presented horizontally, and x
vertically.
. graph hbar (mean) numeric_var, over(cat_var)
x
|
|------+
first group | |
|------+ same conceptual layout:
| numeric_var still appears
|----------+ on y, cat_var on x
second group | |
|----------+
. |
. |
+----------------- y
5 7
The syntax for vertical and horizontal bar charts is the same; all that
is required is changing bar to hbar or hbar to bar.
group_options
over(varname[, over_subopts]) specifies a categorical variable over which
the yvars are to be repeated. varname may be string or numeric. Up
to two over() options may be specified when multiple yvars are
specified, and up to three over()s may be specified when one yvar is
specified; options may be specified; see Examples of syntax and
Multiple over()s under Remarks below.
nofill specifies that missing subcategories be omitted. For instance,
consider
. graph bar (mean) y, over(division) over(region)
Say that one of the divisions has no data for one of the regions,
either because there are no such observations or because y==. for
such observations. In the resulting chart, the bar will be missing:
+-+ +-+ +-+ +-+ +-+
| | | | | | | | | |
| | | | | | | | | |
-------------------------------------------
div_1 div_2 div_3 div_1 div_2 div_3
region_1 region_2
If you specify nofill, the missing category will be removed from the
chart:
+-+ +-+ +-+ +-+ +-+
| | | | | | | | | |
| | | | | | | | | |
------------------------------------
div_1 div_2 div_3 div_1 div_3
region_1 region_2
missing specifies that missing values of the over() variables be kept as
their own categories, one for ., another for .a, etc. The default is
to act as if such observations simply did not appear in the dataset;
the observations are ignored. An over() variable is considered to be
missing if it is numeric and contains a missing value or if it is
string and contains "".
allcategories specifies that all categories in the entire dataset be
retained for the over() variables. When if or in is specified
without allcategories, the graph is drawn, completely excluding any
categories for the over() variables that do not occur in the
specified subsample. With the allcategories option, categories that
do not occur in the subsample still appear in the legend, and
zero-height bars are drawn where these categories would appear. Such
behavior can be convenient when comparing graphs of subsamples that
do not include completely common categories for all over() variables.
This option has an effect only when if or in is specified or if there
are missing values in the variables. allcategories may not be
combined with by().
yvar_options
ascategory specifies that the yvars be treated as the first over() group;
see Treatment of bars under Remarks below. ascategory is a useful
option.
When you specify ascategory, results are the same as if you specified
one yvar and introduced a new first over() variable. Anyplace you
read in the documentation that something is done over the first
over() category, or using the first over() category, it will be done
over or using yvars.
Suppose that you specified
. graph bar y1 y2 y3, ascategory whatever_other_options
The results will be the same as if you typed
. graph bar y, over(newcategoryvariable) whatever_other_options
with a long rather than wide dataset in memory.
asyvars specifies that the first over() group be treated as yvars. See
Treatment of bars under Remarks below.
When you specify asyvars, results are the same as if you removed the
first over() group and introduced multiple yvars. If you previously
had k yvars and, in your first over() category, G groups, results
will be the same as if you specified k*G yvars and removed the
over(). Anyplace you read in the documentation that something is
done over the yvars or using the yvars, it will be done over or using
the first over() group.
Suppose that you specified
. graph bar y, over(group) asyvars whatever_other_options
Results will be the same as if you typed
. graph bar y1 y2 y3 ..., whatever_other_options
with a wide rather than long dataset in memory. Variables y1, y2,
..., are sometimes called the virtual yvars.
percentages specifies that bar heights be based on percentages that
yvar_i represents of all the yvars. That is,
. graph bar (mean) inc_male inc_female
would produce a chart with bar height reflecting average income.
. graph bar (mean) inc_male inc_female, percentage
would produce a chart with the bar heights being
100*inc_male/(inc_male+inc_female) and
100*inc_female/(inc_male+inc_female).
If you have one yvar and want percentages calculated over the first
over() group, specify the asyvars option. For instance,
. graph bar (mean) wage, over(i) over(j)
would produce a chart where bar heights reflect mean wages.
. graph bar (mean) wage, over(i) over(j) asyvars percentages
would produce a chart where bar heights are
100*( mean_ij / (Sum_i mean_ij) )
Option stack is often combined with option percentage.
stack specifies that the yvar bars be stacked.
. graph bar (mean) inc_male inc_female,
over(region) percentage stack
would produce a chart with all bars being the same height, 100%.
Each bar would be two bars stacked (percentage of inc_male and
percentage of inc_female), so the division would show the relative
shares of inc_male and inc_female of total income.
To stack bars over the first over() group, specify the asyvars
option:
. graph bar (mean) wage, over(sex) over(region)
asyvars percentage stack
cw specifies casewise deletion. If cw is specified, observations for
which any of the yvars are missing are ignored. The default is to
calculate the requested statistics by using all the data possible.
lookofbar_options
outergap(*#) and outergap(#) specify the gap between the edge of the
graph to the beginning of the first bar and the end of the last bar
to the edge of the graph.
outergap(*#) specifies that the default be modified. Specifying
outergap(*1.2) increase the gap by 20%, and specifying outergap(*.8)
reduces the gap by 20%.
outergap(#) specifies the gap as a percentage-of-bar-width units.
outergap(50) specifies that the gap be half the bar width.
bargap(#) specifies the gap to be left between yvar bars as a
percentage-of-bar-width units. The default is bargap(0), meaning
that bars touch.
bargap() may be specified as positive or negative numbers.
bargap(10) puts a small gap between the bars (the precise amount
being 10% of the width of the bars). bargap(-30) overlaps the bars
by 30%.
bargap() affects only the yvar bars. If you want to change the gap
for the first, second, or third over() groups, specify the
over_subopt gap() inside the over() itself; see Suboptions for use
with over() and yvaroptions() below.
intensity(#) and intensity(*#) specify the intensity of the color used to
fill the inside of the bar. intensity(#) specifies the intensity,
and intensity(*#) specifies the intensity relative to the default.
By default, the bar is filled with the color of its border,
attenuated. Specify intensity(*#), #<1, to attenuate it more and
specify intensity(*#), #>1, to amplify it.
Specify intensity(0) if you do not want the bar filled at all.
Specify intensity(100) if you want the bar to have the same intensity
as the bar's outline.
lintensity(#) and lintensity(*#) specify the intensity of the line used
to outline the bar. lintensity(#) specifies the intensity, and
lintensity(*#) specifies the intensity relative to the default.
By default, the bar is outlined at the same intensity at which it is
filled or at an amplification of that, which depending on your chosen
scheme; see [G] schemes intro. If you want the bar outlined in the
darkest possible way, specify intensity(255). If you wish simply to
amplify the outline, specify intensity(*#), #>1, and if you wish to
attenuate the outline, specify intensity(*#), #<1.
pcycle(#) specifies how many variables are to be plotted before the
pstyle of the bars for the next variable begins again at the pstyle
of the first variable -- p1bar (with the bars for the variable
following that using p2bar and so). Put another way: # specifies how
quickly the look of bars is recycled when more than # variables are
specified. The default for most schemes is pcycle(15).
bar(#, barlook_options) specifies the look of the yvar bars. bar(1, ...)
refers to the bar associated with the first yvar, bar(2, ...) refers
to the bar associated with the second, and so on. The most useful
barlook_option is color(colorstyle), which sets the color of the bar.
For instance, you might specify bar(1, color(green)) to make the bar
associated with the first yvar green. See [G] colorstyle for a list
of color choices, and see [G] barlook_options for information on the
other barlook_options.
legending_options
legend_option controls the legend. If more than one yvar is specified, a
legend is produced. Otherwise, no legend is needed because the
over() groups are labeled on the categorical x axis. See [G]
legend_option, and see Treatment of bars under Remarks below.
nolabel specifies that, in automatically constructing the legend, the
variable names of the yvars be used in preference to "mean of
varname" or "sum of varname", etc.
yvaroptions(over_subopts) allows you to specify over_subopts for the
yvars. This is seldom done.
showyvars specifies that, in addition to building a legend, the
identities of the yvars be shown on the categorical x axis. If
showyvars is specified, it is typical also to specify legend(off).
blabel() allows you to add labels on top of the bars; see [G]
blabel_option.
axis_options
yalternate and xalternate switch the side on which the axes appear.
Used with graph bar, yalternate moves the numerical y axis from the
left to the right; xalternate moves the categorical x axis from the
bottom to the top.
Used with graph hbar, yalternate moves the numerical y axis from the
bottom to the top; xalternate moves the categorical x axis from the
left to the right.
If your scheme by default puts the axes on the opposite sides, then
yalternate and xalternate reverse their actions.
exclude0 specifies that the numerical y axis need not be scaled to
include 0.
yreverse specifies that the numerical y axis have its scale reversed so
that it runs from maximum to minimum. This option causes bars to
extend down rather than up (graph bar) or from right to left rather
than from left to right (graph hbar).
axis_scale_options specify how the numerical y axis is scaled and how it
looks; see [G] axis_scale_options. There you will also see option
xscale() in addition to yscale(). Ignore xscale(), which is
irrelevant for bar charts.
axis_label_options specify how the numerical y axis is to be labeled.
The axis_label_options also allow you to add and suppress grid lines;
see [G] axis_label_options. There you will see that, in addition to
options ylabel(), ytick(), ..., ymtick(), options xlabel(), ...,
xmtick() are allowed. Ignore the x*() options, which are irrelevant
for bar charts.
ytitle() overrides the default title for the numerical y axis; see [G]
axis_title_options. There you will also find option xtitle()
documented, which is irrelevant for bar charts.
title_and_other options
text() adds text to a specified location on the graph; see [G]
added_text_option. The basic syntax of text() is
text(#_y #_x "text")
text() is documented in terms of twoway graphs. When used with bar
charts, the "numeric" x axis is scaled to run from 0 to 100.
yline() adds horizontal (bar) or vertical (hbar) lines at specified y
values; see [G] added_line_options. The xline() option, also
documented there, is irrelevant for bar charts. If your interest is
in adding grid lines, see [G] axis_label_options.
aspect_option allows you to control the relationship between the height
and width of a graph's plot region; see [G] aspect_option.
std_options allow you to add titles, control the graph size, save the
graph on disk, and much more; see [G] std_options.
by(varlist, ...) draws separate plots within one graph; see [G] by_option
and see Use with by() under Remarks below.
Suboptions for use with over() and yvaroptions()
relabel(# "text" ...) specifies text to override the default category
labeling. Pretend that variable sex took on two values and you typed
. graph bar ..., ... over(sex, relabel(1 "Male" 2 "Female"))
The result would be to relabel the first value of sex to be "Male"
and the second value, "Female"; "Male" and "Female" would appear on
the categorical x axis to label the bars. This would be the result,
regardless of whether variable sex were string or numeric and
regardless of the codes actually stored in the variable to record
sex.
That is, # refers to category number, which is determined by sorting
the unique values of the variable (here sex) and assigning 1 to the
first value, 2 to the second, and so on. If you are unsure as to
what that ordering would be, the easy way to find out is to type
. tabulate sex
If you also plan on specifying graph bar's or graph hbar's missing
option,
. graph bar ..., ... missing over(sex, relabel(...))
then type
. tabulate sex, missing
to determine the coding. See [R] tabulate oneway.
Relabeling the values does not change the order in which the bars are
displayed.
You may create multiple-line labels by using quoted strings within
quoted strings:
over(varname, relabel(1 `" "Male" "patients" "'
2 `" "Female" "patients" "'))
When specifying quoted strings within quoted strings, remember to use
compound double quotes `" and "' on the outer level.
relabel() may also be specified inside yvaroptions(). By default,
the identity of the yvars is revealed in the legend, so specifying
yvaroptions(relabel()) changes the legend. Because it is the legend
that is changed, using legend(label()) is preferred; see
legending_options above. In any case, specifying
yvaroptions(relabel(1 "Males" 2 "Females"))
changes the text that appears in the legend for the first yvar and
the second yvar. # in relabel(# ...) refers to yvar number. Here
you may not use the nested quotes to create multiline labels; use the
legend(label()) option because it provides multiline capabilities.
label(cat_axis_label_options) determines other aspects of the look of the
category labels on the x axis. Except for label(labcolor()) and
label(labsize()), these options are seldom specified; see [G]
cat_axis_label_options.
axis(cat_axis_line_options) specifies how the axis line is rendered.
This is a seldom specified option. See [G] cat_axis_line_options.
gap(#) and gap(*#) specify the gap between the bars in this over() group.
gap(#) is specified in percentage-of-bar-width units, so gap(67)
means two-thirds the width of a bar. gap(*#) allows modifying the
default gap. gap(*1.2) would increase the gap by 20%, and gap(*.8)
would decrease the gap by 20%.
To understand the distinction between over(..., gap()) and option
bargap(), consider
. graph bar revenue profit, bargap(...) over(division, gap(...))
bargap() sets the distance between the revenue and profit bars.
over(,gap()) sets the distance between the bars for the first
division and the second division, the second division and the third,
and so on. Similarly, in
. graph bar revenue profit, bargap(...)
over(division, gap(...))
over(year, gap(...))
over(division, gap()) sets the gap between divisions and over(year,
gap()) sets the gap between years.
sort(varname), sort(#), and sort((stat) varname) control how bars are
ordered. See How bars are ordered and Reordering the bars under
Remarks below.
sort(varname) puts the bars in the order of varname; see Putting the
bars in a prespecified order under Remarks below.
sort(#) puts the bars in height order. # refers to the yvar number
on which the ordering should be performed; see Putting the bars in
height order under Remarks below.
sort((stat) varname) puts the bars in an order based on a calculated
statistic; see Putting the bars in a derived order under Remarks
below.
descending specifies that the order of the bars -- default or as
specified by sort() -- be reversed.
reverse specifies that the categorical scale run from maximum to minimum
rather than the default minimum to maximum. Among other things, when
combined with bargap(-#), reverse causes the sequence of overlapping
to be reversed.
Remarks
Remarks are presented under the following headings:
Introduction
Examples of syntax
Treatment of bars
Treatment of data
Multiple bars (overlapping the bars)
Controlling the text of the legend
Multiple over()s (repeating the bars)
Nested over()s
Charts with many categories
How bars are ordered
Reordering the bars
Putting the bars in a prespecified order
Putting the bars in height order
Putting the bars in a derived order
Reordering the bars, example
Use with by()
History
Introduction
Let us show you some bar charts:
. sysuse citytemp
. graph bar (mean) tempjuly tempjan, over(region)
bargap(-30)
legend( label(1 "July") label(2 "January") )
ytitle("Degrees Fahrenheit")
title("Average July and January temperatures")
subtitle("by regions of the United States")
note("Source: U.S. Census Bureau, U.S. Dept. of Commerce")
(click to run)
. sysuse citytemp, clear
. graph hbar (mean) tempjan, over(division) over(region) nofill
ytitle("Degrees Fahrenheit")
title("Average January temperature")
subtitle("by region and division of the United States")
note("Source: U.S. Census Bureau, U.S. Dept. of Commerce")
(click to run)
. sysuse nlsw88, clear
. graph bar (mean) wage, over(smsa) over(married) over(collgrad)
title("Average Hourly Wage, 1988, Women Aged 34-46")
subtitle("by College Graduation, Marital Status,
and SMSA residence")
note("Source: 1988 data from NLS, U.S. Dept. of Labor,
Bureau of Labor Statistics")
(click to run)
. sysuse educ99gdp, clear
. generate total = private + public
. graph hbar (asis) public private,
over(country, sort(total) descending) stack
title( "Spending on tertiary education as % of GDP,
1999", span pos(11) )
subtitle(" ")
note("Source: OECD, Education at a Glance 2002", span)
(click to run)
In the sections that follow, we explain how each of the above graphs --
and others -- are produced.
Examples of syntax
Below we show you some graph bar commands and tell you what each would
do:
graph bar revenue
One big bar showing average revenue.
graph bar revenue profit
Two bars, one showing average revenue and the other showing
average profit.
graph bar revenue, over(division)
#_of_divisions bars showing average revenue for each division.
graph bar revenue profit, over(division)
2*#_of_divisions bars showing average revenue and average
profit for each division. The grouping would look like this
(assuming three divisions):
+-+ +-+ +-+
| | | | | |
| |-+ | |-+ | |-+
| | | | | | | | |
----------------------------
division division division
graph bar revenue, over(division) over(year)
#_of_divisions*#_of_years bars showing average revenue for each
division, repeated for each of the years. The grouping would
look like this (assuming three divisions and 2 years):
+-+ +-+ +-+ +-+ +-+ +-+
| | | | | | | | | | | |
| | | | | | | | | | | |
-----------------------------------------------------------
division division division division division division
year year
graph bar revenue, over(year) over(division)
same as above but ordered differently. In the previous example
we typed over(division) over(year). This time, we reverse it:
+-+ +-+ +-+ +-+ +-+ +-+
| | | | | | | | | | | |
| | | | | | | | | | | |
--------------------------------------------------
year year year year year year
division division division
graph bar revenue profit, over(division) over(year)
2*#_of_divisions*#_of_years bars showing average revenue and
average profit for each division, repeated for each of the
years. The grouping would look like this (assuming three
divisions and 2 years):
+-+ +-+ +-+ +-+ +-+ +-+
| | | | | | | | | | | |
| |-+ | |-+ | |-+ | |-+ | |-+ | |-+
| | | | | | | | | | | | | | | | | |
-----------------------------------------------------------
division division division division division division
year year
graph bar (sum) revenue profit, over(division) over(year)
2*#_of_divisions*#_of_years bars showing the sum of revenue and
sum of profit for each division, repeated for each of the
years.
graph bar (median) revenue profit, over(division) over(year)
2*#_of_divisions*#_of_years bars showing the median of revenue
and median of profit for each division, repeated for each of
the years.
graph bar (median) revenue (mean) profit, over(division) over(year)
2*#_of_divisions*#_of_years bars showing the median of revenue
and mean of profit for each division, repeated for each of the
years.
Treatment of bars
Assume someone tells you that the average January temperature in the
Northeast of the United States is 27.9 degrees Fahrenheit, 27.1 degrees
in the North Central, 46.1 in the South, and 46.2 in the West. You could
enter these statistics and draw a bar chart:
. input ne nc south west
ne nc south west
1. 27.9 21.7 46.1 46.2
2. end
. graph bar (asis) ne nc south west
(click to run)
The above is admittedly not a great-looking chart, but specifying a few
options could fix that. The important thing to see right now is that,
when we specify multiple yvars, (1) the bars touch, (2) the bars are
different colors (or at least different shades of gray), and (3) the
meaning of the bars is revealed in the legend.
We could enter this data another way:
. clear
. input str10 region float tempjan
region tempjan
1. N.E. 27.9
2. "N. Central" 21.7
3. South 46.1
4. West 46.2
5. end
. graph bar (asis) tempjan, over(region)
(click to run)
Observe that, when we generate multiple bars via an over() option, (1)
the bars do not touch, (2) the bars are all the same color, and (3) the
meaning of the bars is revealed by how the categorical x axis is labeled.
These differences in the treatment of the bars in the multiple yvars case
and the over() case are general properties of graph bar and graph hbar:
multiple yvars over() groups
-------------------------------------------------------------
bars touch yes no
bars different colors yes no
bars identified via ... legend axis label
-------------------------------------------------------------
Option ascategory causes multiple yvars to be presented as if they were
over() groups, and option asyvars causes over() groups to be presented as
if they were multiple yvars. Thus
. graph bar (asis) tempjan, over(region)
would produce the first chart and
. graph bar (asis) ne nc south west, ascategory
would produce the second.
Treatment of data
In the previous two examples, we already had the statistics we wanted to
plot: 27.9 (Northeast), 21.7 (North Central), 46.1 (South), and 46.2
(West). We entered the data, and we typed
. graph bar (asis) ne nc south west
or
. graph bar (asis) tempjan, over(region)
We do not have to know the statistics ahead of time: graph bar and graph
hbar can calculate statistics for us. If we had datasets with lots of
observations (say, cities of the United States), we could type
. graph bar (mean) ne nc south west
or
. graph bar (mean) tempjan, over(region)
and obtain the same graphs. All we need do is change (asis) to (mean).
In the first example, the data would be organized the wide way:
cityname ne nc south west
-----------------------------------------------
name of city 42 . . .
another city . 28 . .
...
-----------------------------------------------
and, in the second example, the data would be organized the long way:
cityname region tempjan
-------------------------------------
name of city ne 42
another city nc 28
...
-------------------------------------
We have such a dataset, organized the long way. In citytemp.dta, we have
information on 956 U.S. cities, including the region in which each is
located and its average January temperature:
. sysuse citytemp, clear
. list region tempjan if _n < 3 | _n > 954
+------------------+
| region tempjan |
|------------------|
1. | NE 16.6 |
2. | NE 18.2 |
955. | West 72.6 |
956. | West 72.6 |
+------------------+
With this data, we can type
. graph bar (mean) tempjan, over(region)
(click to run)
We just produced the same bar chart we previously produced when we
entered the statistics 27.9 (Northeast), 21.7 (North Central), 46.1
(South), and 46.2 (West) and typed
. graph bar (asis) tempjan, over(region)
When we do not specify (asis) or (mean) (or (median) or (sum) or (p1) or
any of the other stats allowed), (mean) is assumed. Thus (...) is often
omitted when (mean) is desired, and we could have drawn the previous
graph by typing
. graph bar tempjan, over(region)
Some users even omit typing (...) in the (asis) case because calculating
the mean of one observation results in the number itself. Thus in the
previous section, rather than typing
. graph bar (asis) ne nc south west
and
. graph bar (asis) tempjan, over(region)
We could have typed
. graph bar ne nc south west
and
. graph bar tempjan, over(region)
Multiple bars (overlapping the bars)
In citytemp.dta, in addition to variable tempjan, there is variable
tempjuly, which the average July temperature. We can include both
averages in one chart, by region:
. sysuse citytemp, clear
. graph bar (mean) tempjuly tempjan, over(region)
(click to run)
We can improve the look of the chart by
1. including the legend_option legend(label()) to change the text of
the legend; see [G] legend_option;
2. including the axis_title_option ytitle() to add a title saying
"Degrees Fahrenheit"; see [G] axis_title_options;
3. including the title_options title(), subtitle(), and note() to
say what the graph is about and from where the data came; see [G]
title_options.
Doing all that produces
. graph bar (mean) tempjuly tempjan, over(region)
legend( label(1 "July") label(2 "January") )
ytitle("Degrees Fahrenheit")
title("Average July and January temperatures")
subtitle("by regions of the United States")
note("Source: U.S. Census Bureau, U.S. Dept. of Commerce")
(click to run)
We can make one more improvement to this chart by overlapping the bars.
Below we add the option bargap(-30):
. graph bar (mean) tempjuly tempjan, over(region)
new -> bargap(-30)
legend( label(1 "July") label(2 "January") )
ytitle("Degrees Fahrenheit")
title("Average July and January temperatures")
subtitle("by regions of the United States")
note("Source: U.S. Census Bureau, U.S. Dept. of Commerce")
(click to run)
bargap(#) specifies the distance between the yvar bars (i.e., between the
bars for tempjuly and tempjan); # is in percentage-of-bar-width units, so
barwidth(-30) means that the bars overlap by 30%. bargap() may be
positive or negative; its default is 0.
Controlling the text of the legend
In the above example, we changed the text of the legend by specifying the
legend option:
legend( label(1 "July") label(2 "January") )
We could just as well have changed the text of the legend by typing
yvaroptions( relabel(1 "July" 2 "January") )
Which you use makes no difference, but we prefer legend(label()) to
yvaroptions(relabel()) because legend(label()) is the way to modify the
contents of a legend in a twoway graph; so why do bar charts differently?
Multiple over()s (repeating the bars)
Option over(varname) repeats the yvar bars for each unique value of
varname. Using citytemp.dta, if we typed
. graph bar (mean) tempjuly tempjan
we would obtain two (fat) bars. When we type
. graph bar (mean) tempjuly tempjan, over(region)
we obtain two (thinner) bars for each of the four regions. (We typed
exactly this command in Multiple bars above.)
You may repeat the over() option. You may specify over() twice when you
specify two or more yvars and up to three times when you specify just one
yvar.
In dataset nlsw88.dta, we have information on 2,246 women:
. sysuse nlsw88, clear
. graph bar (mean) wage, over(smsa) over(married) over(collgrad)
title("Average Hourly Wage, 1988, Women Aged 34-46")
subtitle("by College Graduation, Marital Status,
and SMSA residence")
note("Source: 1988 data from NLS, U.S. Dept. of Labor,
Bureau of Labor Statistics")
(click to run)
If you strip away the title_options, the above command reads
. graph bar (mean) wage, over(smsa) over(married) over(collgrad)
In this three-over() case, the first over() is treated as multiple yvars:
the bars touch, the bars are assigned different colors, and the meaning
of the bars is revealed in the legend. When you specify three over()
groups, the first is treated the same way as multiple yvars. This means
that if we wanted to separate the bars, we could specify option
bargap(#), #>0, and if we wanted them to overlap, we could specify
bargap(#), #<0.
Nested over()s
Sometimes you have multiple over() groups with one group explicitly
nested within the other. In the citytemp.dta dataset, we have variables
region and division, and division is nested within region. The Census
Bureau divides the United States into four regions and into nine
divisions, which work like this
-----------------------------------------------------
Region Division
-----------------------------------------------------
1. Northeast 1. New England
2. Mid Atlantic
-----------------------------------------------------
2. North Central 3. East North Central
4. West North Central
-----------------------------------------------------
3. South 5. South Atlantic
6. East South Central
7. West South Central
-----------------------------------------------------
4. West 8. Mountain
9. Pacific
-----------------------------------------------------
Were we to type
. graph bar (mean) tempjuly tempjan, over(division) over(region)
we would obtain a chart with space allocated for 9*4 = 36 groups, of
which only nine would be used:
+-+ +-+ +-+
| | | | | |
| |-+ | |-+ | |-+
| | | | | | | | |
-------------------------------------------------------------------
1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 ... 1 2 3 4 5 6 7 8 9
region 1 region 2 region 4
The nofill option prevents the chart from including the unused
categories:
. sysuse citytemp, clear
. graph bar tempjuly tempjan, over(division) over(region) nofill
bargap(-30)
ytitle("Degrees Fahrenheit")
legend( label(1 "July") label(2 "January") )
title("Average July and January temperatures")
subtitle("by region and division of the United States")
note("Source: U.S. Census Bureau, U.S. Dept. of Commerce")
(click to run)
The above chart, if we omit one of the temperatures, also looks good
horizontally:
. graph hbar (mean) tempjan, over(division) over(region) nofill
ytitle("Degrees Fahrenheit")
title("Average January temperature")
subtitle("by region and division of the United States")
note("Source: U.S. Census Bureau, U.S. Dept. of Commerce")
(click to run)
Charts with many categories
Using nlsw88.dta, we want to draw the chart
. sysuse nlsw88
. graph bar wage, over(industry) over(collgrad)
Variable industry records industry of employment in 12 categories, and
variable collgrad records whether the woman is a college graduate. Thus
we will have 24 bars. We draw the above and quickly discover that the
long labels associated with industry result in much overprinting along
the horizontal x axis.
Horizontal bar charts work better than vertical bar charts when labels
are long. We change our command to read
. graph hbar wage, over(ind) over(collgrad)
That works better, but now we have overprinting problems of a different
sort: the letters of one line are touching the letters of the next.
Graphs are by default 4x5: 4 inches tall by 5 inches wide. Here we need
to make the chart taller, and that is the job of the region_option
ysize(). Below we make a chart that is 7 inches tall:
. sysuse nlsw88, clear
. graph hbar wage, over(ind, sort(1)) over(collgrad)
title("Average hourly wage, 1988, women aged 34-46", span)
subtitle(" ")
note("Source: 1988 data from NLS, U.S. Dept. of Labor,
Bureau of Labor Statistics", span)
ysize(7)
(click to run)
The important option in the above is ysize(7), which made the graph
taller than usual; see [G] region_options. Concerning the other options:
over(ind, sort(1)) over(collgrad)
sort(1) is specified so that the bars would be sorted on mean
wage. The 1 says to sort on the first yvar; see Reordering the
bars below.
title("Average hourly wage, 1988, women aged 34-46", span)
span is specified so that the title, rather than being centered
over the plot region, would be centered over the entire graph.
Here the plot region (the part of the graph where the real chart
appears, ignoring the labels) is narrow, and centering over that
was not going to work. See [G] region_options for a description
of the graph region and plot region, and see [G] title_options
and [G] textbox_options for a description of span.
subtitle(" ")
We specified this because the title looked too close to the graph
without it. We could have done things properly and specified a
margin() suboption within the title(), but we often find it
easier to include a blank subtitle. We typed subtitle(" ") and
not subtitle(""). We had to include the blank, or the subtitle
would not have appeared.
note("Source: 1988 data from NLS, ...", span)
span is specified so that the note would be left-justified in the
graph rather than just in the plot region.
How bars are ordered
The default is to place the bars in the order of the yvars and to order
each set of over(varname) groups according to the values of varname. Let
us consider some examples:
graph bar (sum) revenue profit
Bars appear in the order specified: revenue and profit.
graph bar (sum) revenue, over(division)
Bars are ordered according to the values of variable division.
If division is a numeric variable, the lowest division number comes
first, followed by the next lowest, and so on. This is true even if
variable division has a value label. Say that division 1 has been
labeled "Sales" and division 2 is labeled "Development". The bars
will be in the order Sales followed by Development.
If division is a string variable, the bars will be ordered by the
sort order of the values of division (meaning alphabetically, but
with capital letters placed before lowercase letters). If variable
division contains the values "Sales" and "Development", the bars will
be in the order Development followed by Sales.
graph bar (sum) revenue profit, over(division)
Bars appear in the order specified, revenue and profit, and are
repeated for each division, which will be ordered as explained above.
graph bar (sum) revenue, over(division) over(year)
Bars appear ordered by the values of division, as previously
explained, and then that is repeated for each of the years. The
years are ordered according to the values of the variable year,
following the same rules as applied to the variable division.
graph bar (sum) revenue profit, over(division) over(year)
Bars appear in the order specified, profit and revenue, repeated for
division ordered on the values of variable division, repeated for
year ordered on the values of variable year.
Reordering the bars
There are three ways to reorder the bars:
1. You want to control the order in which the elements of each
over() group appear. Your divisions might be named Development,
Marketing, Research, and Sales, alphabetically speaking, but you
want them to appear in the more logical order Research,
Development, Marketing, and Sales.
2. You wish to order the bars according to their heights. You wish
to draw the graph
. graph bar (sum) empcost, over(division)
and you want the divisions ordered by total employee cost.
3. You wish to order on some other derived value.
We will consider each of these desires separately.
Putting the bars in a prespecified order
We have drawn the graph
. graph (sum) bar empcost, over(division)
Variable division is a string containing "Development", "Marketing",
"Research", and "Sales". We want to draw the chart, placing the
divisions in the order Research, Development, Marketing, and Sales.
To do that, we create a new numeric variable that orders division as we
would like:
. generate order = 1 if division=="Research"
. replace order = 2 if division=="Development"
. replace order = 3 if division=="Marketing"
. replace order = 4 if division=="Sales"
We can name the variable and create it however we wish, but we must be
sure that there is a one-to-one correspondence between the new variable
and the over() group's values. We then specify the over()'s
sort(varname) option:
. graph bar (sum) empcost, over( division, sort(order) )
If you want to reverse the order, you may specify the descending
suboption:
. graph bar (sum) empcost, over(division, sort(order) descending)
Putting the bars in height order
We have drawn the graph
. graph bar (sum) empcost, over(division)
and now wish to put the bars in height order, shortest first. We type
. graph bar (sum) empcost, over( division, sort(1) )
If we wanted the tallest first, type
. graph bar empcost, over(division, sort(1) descending)
The 1 in sort(1) refers to the first (and here only) yvar. If we had
multiple yvars, we might type
. graph bar (sum) empcost othcost, over( division, sort(1) )
and we would have a chart showing employee cost and other cost, sorted on
employee cost. If we typed
. graph bar (sum) empcost othcost, over( division, sort(2) )
the graph would be sorted on other cost.
We can use sort(#) on the second over() group as well:
. graph bar (sum) empcost, over( division, sort(1) )
over( country, sort(1) )
Country will be ordered on the sum of the heights of the bars.
Putting the bars in a derived order
We have employee cost broken into two categories: empcost_direct and
empcost_indirect. Variable emp_cost is the sum of the two. We wish to
make a chart showing the two costs, stacked, over division, and we want
the bars ordered on the total height of the stacked bars. We type
. graph bar (sum) empcost_direct empcost_indirect,
stack
over(division, sort((sum) empcost) descending)
Reordering the bars, example
We have a dataset showing the spending on tertiary education as a percent
of GDP from the 2002 edition of Education at a Glance: OECD Indicators
2002:
. sysuse educ99gdp, clear
. list
+----------------------------------+
| country public private |
|----------------------------------|
1. | Australia .7 .7 |
2. | Britain .7 .4 |
3. | Canada 1.5 .9 |
4. | Denmark 1.5 .1 |
5. | France .9 .4 |
|----------------------------------|
6. | Germany .9 .2 |
7. | Ireland 1.1 .3 |
8. | Netherlands 1 .4 |
9. | Sweden 1.5 .2 |
10. | United States 1.1 1.2 |
+----------------------------------+
We wish to graph total spending on education and simultaneously show the
distribution of that total between public and private expenditures. We
want the bar sorted on total expenditures:
. generate total = private + public
. graph hbar (asis) public private,
over(country, sort(total) descending) stack
title( "Spending on tertiary education as % of GDP, 1999",
span pos(11) )
subtitle(" ")
note("Source: OECD, Education at a Glance 2002", span)
(click to run)
Or perhaps we wish to disguise the total expenditures and focus the graph
exclusively on the share of spending that is public and private:
. generate frac = private/(private + public)
. graph hbar (asis) public private,
over(country, sort(frac) descending) stack percent
title("Public and private spending on tertiary education, 1999"
> ,
span pos(11) )
subtitle(" ")
note("Source: OECD, Education at a Glance 2002", span)
(click to run)
The only differences between the two graph hbar commands are as follows:
1. The percentage option was added to change the yvars public and
private from spending amounts to percent each is of the total.
2. The order of the bars was changed.
3. The title was changed.
Use with by()
graph bar and graph hbar may be used with by(), but in general, you want
to use over() in preference to by(). Bar charts are explicitly
categorical and do an excellent job of presenting summary statistics for
multiple groups in one chart.
A good use of by(), however, is when you are ordering the bars and you
wish to emphasize that the ordering is different for different groups.
For instance,
. sysuse nlsw88, clear
. graph hbar wage, over(occ, sort(1)) by(union)
(click to run)
The above graph orders the bars by height (hourly wage); the orderings
are different for union and nonunion workers.
History
The first published bar chart appeared in William Playfair's Commercial
and Political Atlas (1786). See Tufte (1983, 32-33) or Beniger and Robyn
(1978) for more historical information.
Also see
Manual: [G] graph bar
Help: [G] graph dot; [D] collapse, [R] table
